Data Engineer Professional cheat sheet
Last-minute reference for the Databricks Data Engineer Professional exam. This one tests production judgement, so most answers concern what survives contact with reality.
Exam facts
| Questions | 60 scored (outline from 9 October 2026) |
| Duration | 120 minutes |
| Fee | 200 USD |
| Validity | 2 years |
| Passing score | Not published by Databricks |
Section weights
| Section | Weight |
|---|---|
| Developing code for data processing | 23% |
| Cost & performance optimisation | 15% |
| Data ingestion & acquisition (incl. sharing, federation) | 12% |
| Data manipulation | 12% |
| Monitoring and alerting | 10% |
| Debugging and deploying | 10% |
| Security and compliance | 8% |
| Data governance | 5% |
| Data modelling | 5% |
Production patterns
| Requirement | Answer |
|---|---|
| Exactly-once effect across restarts | Checkpointed offsets + idempotent or transactional sink |
Append inside foreachBatch duplicated after restart | foreachBatch is at-least-once; use batch ID or MERGE |
| Streaming aggregation in append mode | Needs a watermark; emits once the window closes |
| Python UDF dominating runtime on 400M rows | Built-in expressions or a vectorised UDF |
| Heavy spill joining 4 TB to 30 MB | Check why the small side is not broadcast |
| Identical behaviour for backfill and stream | One transformation, parameterised by read/write mode |
| Logic must be testable outside a notebook | Versioned module with unit tests |
| Cost tripled, data grew 20%, heavy listing | Small file accumulation → compact |
| Frequent filters on two high-cardinality columns | Liquid Clustering, not partition |
| Managed table, no one wants to pick layout keys | CLUSTER BY AUTO with Predictive Optimization |
| Many small updates rewriting large files | Deletion vectors |
| Downstream needs only changed rows, incl. deletes | Change Data Feed |
| JSON with a shifting shape | VARIANT + parse_json / variant_get |
| Streaming job silently behind at peak | Input backlog / source lag |
| Alert only when actionable | Alert on final failure after retries |
Security and governance
- Erasure request on an append-only table → retained versions and downstream copies must be addressed within the retention policy; with deletion vectors, rewrite the files (
REORG ... APPLY (PURGE)) andVACUUM. Deleting the current row is not enough. - Mask every column with one governed tag, across all tables → ABAC policy on the tag, not per-table masks.
- SELECT granted, query still denied → missing
USE CATALOG/USE SCHEMAon the parents. - Service principal with workspace admin → excessive privilege; scope to what the job needs.
- Impact of a schema change → lineage in the governance catalog, not a code search.
- Make tables findable → tags and comments on the securable.
- Divergent metric definitions → one governed definition, such as a Unity Catalog metric view, both consumers use.
Ingestion
- CDC feed → ordered merge including deletes and late or out-of-order events
- CDC from SQL Server, MySQL or PostgreSQL → Lakeflow Connect
- Partner needs live data, no workspace accounts → OpenSharing (formerly Delta Sharing); joint analysis without exposing raw data → Clean Rooms
- Query an external database without copying → Lakehouse Federation
- Rate-limited API with partial pages → backoff, resumable pagination, completeness validation
The published-score trap
No passing score is published. At professional level the margin is narrower, so calibrate on a weighted practice test.
Night-before checklist
- Nine sections since 9 October 2026; code development is 23%
- Exactly-once = checkpoint + idempotent sink
- Liquid Clustering versus partition for high cardinality
- Check ID and proctoring rules — see exam day
Take the 20-question practice test.